In David’s QEX CESSB article, he mentions several times of the value of baseband audio processing applied before his algorithm:
“It is possible to overdrive the algorithm by driving the baseband “RF clipper” into a few dB of clipping beyond what is necessary to remove the Hilbert transform overshoots. This will provide a further increase in average power. I think it is better to do baseband audio processing, however, and let the CESSB system only remove overshoots. Historically, baseband audio processing was not considered particularly effective for increasing average SSB power — because of envelope overshoots. This is no longer true, with the introduction of CESSB. With baseband audio processing, sophisticated multiband compression and clipping is possible, with better results than a simple single-band RF clipper.”
“The processing used does not produce any significant audible artifacts. It may be used in conjunction with speech processing.”
“The system only affects transients that would produce overshoot. Audio compression to increase density, compression, peak limiting, and so on should be applied before the audio input to Figure 11.”
This raises the following questions for the 6000 series implementation:
1. What baseband audio processing is applied before the CESSB algorithm?
2. What does the NOR-DX-DX+ control do? Is it simply overdriving the algorithm as mentioned above?
Bruene, W., "Notes on Speech Clipping and Filtering," QST, Mar., 1952, pp. 54-56;
Clegg, E.T. & Squires, W.K., "Speech Clipping for Single Sideband," QST, July, 1965, pp. 11-15;
Collins, H., "Ordinary and Processed Speech in SSB Applications," QST, Jan., 1969, pp. 17-22;
Kirkwood, B., "Principles of Speech Processing," Ham Radio, Feb., 1975, pp. 18-34;
Moxon, L., "Performance of R.F. Speech Clippers," Ham Radio, Nov. 1972, p. 22;
Sabin, W., "R.F. Clippers for SSB," QST, July, 1967, pp. 13-18;
Tong, D.A., "Distortion Levels in R.F. Clipping," Wireless World, Oct., 1976, pp. 77-81.
The NOR-DX-DX+ control adjusts the level of processing applied to the audio. The further to the right you slide the control, the more processing is used and potentially the more distortion. To the folks we tested on, it was difficult to tell anything was different other than being a lot louder in the NOR/DX range. DX+ added more volume and altered frequency response, also some distortion, although very tolerable. If you're trying to bust a pileup and can't get through on DX, try DX+ -- you might not get a great report on audio quality, but hopefully the added power and intelligibility will help you make the contact.
Thank you for the comments about the voice processing. The reason I was asking about the specific baseband processing applied before CESSB is because I am considering buying an outboard processor and would like to know if applying single or multiband compression and other processing common to the hardware voice channel strips would be of benefit or would be redundant to what is already inside of the 6000 series.
Also, I am pondering what the Waveform API could be put to use for in the area of additional baseband audio processing.
If this information is considered proprietary, I understand. Otherwise, it would be helpful to get some more
Thanks and 73, Bob, N7ZO
We also do various audio processing required to make sure our other pieces work well. As I mentioned, the audio chain is rather complex and it's one of those deals where you could see a high level block diagram, but then you'd ask a question about something and we'd have to open up the block, etc. It is a "devil is in the details" kind of thing.
I would probably use an outboard processor on the MIC before I would use the waveform API. The waveform API is really intended to support new modulation schemes, but it certainly could be used -- you could take it and jack it into an audio plug-in tool on Windows and do all kinds of fancy audio stuff. You just need to be sure to modulate the signal into I/Q before sending it out of the software -- we expect IQ data out, not real stereo.
I'll ping Mark to come read this thread and add his two cents.
I am an audiophile, but I am not a frustrated broadcaster looking for ESSB tones. I am interested in getting the loudest, most copyable signal, which, I would think, is an interest FlexRadio shares.
I tend towards audio that might be a bit sharp for rag-chewing, but punches through to the DX stations. So, for instance, a de-esser might be good for broadcast radio, but it may actually work against intelligibility (as long as there is a clipper downstream). And I doubt that there is a significant effective difference between a parametric EQ and the 8 fixed-band graphic EQ in the 6000 series for the ham radio application.
Two areas that are of interest for a loud intelligible signal are compression and voice symmetry.
Compression: CESSB naturally causes some compression and overdriving it causes more. Can a single or multi-band baseband compressor add more average power to the signal? (Dave, W9GR, seems to recommend it.) Would a multi-band compressor be more effective in front of CESSB? (A multi-band compressor has the advantage of also being an adaptive EQ.)
Voice symmetry: Many of the commercial voice channel strips use all-pass phase shifters to smear the voice peaks and reduce the natural high crest factor of the human voice significantly.
I have here a Symetrix 528E voice processor (with a single
band compressor and voice symmetry correction, among other things) and am trying to get my hands on
a TC Electronic Triple C multiband compressor, but it seems a shame to use
external hardware for this. There are a
number of good software based compressors for the PC that could be patched into the
audio chain for experimental purposes. But the best solutions would be internal to the radio.
Anyways, these are interesting questions and I appreciate your comments on the 6000 series audio chain (and look forward to Mark’s input).
73, Bob, N7ZO
I spent the last 30 years working in broadcast radio and have seen a lot of technology improvements during that time. One of the areas that just keeps getting better is the art of audio processing. In broadcast radio we process the microphone audio to sound the way we want as well as bring up the average output levels with a microphone audio processor. This mic processor is not the final processing on the sound of a broadcast station but rather a key part of it where talent voices are concerned. It is responsible for that larger than life sound you typically hear when the DJ opens the mic.
As Steve mentioned, the Vorsis M1 is a DSP based unit that allows the user to save and recall profiles based on their needs. While it is designed for use in a broadcast facility or recording studio, it is certainly right at home in amateur radio as well. By using the parametric equalizer to tailor the audio, I am able to achieve the sound I want for the job at hand. One profile might be wide and open for ESSB while another could be more narrow and punchy for working DX. By letting the Vorsis process my microphone audio I am able to maintain tighter control over the sound of my signal. I use a Neumann TLM-103 Condenser microphone on my station.
Greg K5GJ and I will be making some audio recordings of the Vorsis in action today. We will record the performance both with CESSB enabled and disabled.
73 de na6m
CESSB behaves like an ideal RF clipper, with minimal distortion and canting. The DX and DX+ settings drive CESSB harder and that does increase talk power but with some distortion. On the other hand, if we drive the rig with a more compressed and EQ'ed voice signal, CESSB may provide additional average power at low distortion levels.
73, Alan K2WS
Now I am not the only one playing with duty cycle and "LOUD Audio" but there is a difference between increasing the duty cycle hence the output and just crappy audio coming out of the transmitter.
Since Flex has marked these 3 levels normal,DX, DX+ one would presume that the efficiency can be seen while running DX . And with that usage the only problem is the majority of the DX stations out there are not listening 4.0 wide or 3.5 wide. Most are restricting their bandwidth to 2.8 or less . (The majority of stations out there capability.)
SO the ESSB venue cannot be the issue with Bulk compression.
If you wish higher duty cycle or "LOUD audio" this is not cleanly accomplished by that slider.
Compression should be added in various spots within the audio spectrum. Not large doses but literally small amounts and the sum total of these increases is what makes your audio "Loud" un distorted, clear and clean. Improved duty cycle .
I have only found multi channel electronic audio processing capable of this feat
To really get ahead of the curve there are a few more things to consider .
Keeping your transmitter not at the full output but far below this level. Where it has overhead . The IMD mix is far less there . Is necessary
If so inclined must keep your amplifier totally linear and not pushed to the wall. . If you wish to run full legal that amplifier better be capable of far in excess of this 1.5 KW level. Again Overhead.
IF 1.5 KW out is your goal better get the big boy pants on amp Not for the watts out but for that reserve capacity
And Yes you can achieve full bodied audio . Fantastic duty cycle running 1.5KW and still be 3DB Better the rest.
He came back with a pream which had a tube and solid state amplifier, you move from solid state amplification to using the tube which gives you a warmth. Also the unit he chose has a compressor and limiter and a few other items in a very small footprint. Of course you have the ability to adjust gain before going into the balance input on the back of the flex. I spent 95.00 and it was the best 95.00 I have ever spent in ham radio. Ido my eq with the flex, do not use the 20 db boost, set mic gain at 10 and then set my tube preamp to desired amount of warmth and gain to what I need to get audio where it is recommended. I use dx1 all the time. Now I get great audio reports, no distortion and with prophiles I have a rag chew pr40 and a dx pr40 and a obnoxious pile buster dx2 pr40. I contribute the advise Time gave me concerning the dsp gain in getting the results I have obtained. And of course my friend and his knowledge in nudging me on some things to do without sitting down to a full fledged sound board.
As mentioned by Bob, N7ZO, a phase rotator is very effective at increasing the average power in speech. When you examine a vocalized phrase you will typically see a highly asymmetric waveform with a high peak to average ratio, or crest factor. Radio stations use the Symmetrix processors to alter the phase relationship between harmonics with the result that the crest factor is reduced substantially. And that permits downstream compressors and limiters to do a more effective job while maintaining high average output power.
Such processing is really nothing more than a multi-pole (4 to 12 poles) all pass filter, typically tuned down around 300 Hz. It does not introduce any unwanted transient behavior, and the human ear cannot discern the difference between processed and unprocessed sound. We are insensitive to the relative phase between harmonics in sounds, whether musical or speech.
A Symmetrix box is pretty expensive for what it does. But if you have any coding savvy then it is really easy to produce All-Pass Filters in software for use in a DAW (digital audio workstation) running between your mic and the transmitter. E.g., VST plugins for Reaper.
73 de Dave, N7AIG
I went back and re-studied David's paper on CESSB in QEX. As a physicist, I know there is no free lunch. While he shows a measurement at the end of the paper showing higher effective average output power under CESSB compared to a "fast acting ALC", that only proves that CESSB is more effective at packing distortion product power into the pass band than this particular ALC.
Once you have a full amplitude signal coming in, neither one can produce stronger actual voice components, but rather, have different ways of generating and packing in additional distortion products into the passband.
No details were provided about how the ALC really behaves, and another design might exhibit entirely different character. ALC is attempting to lower the input drive to control output power to control RF IMD generation, and its recovery time might be a tad slow. CESSB appears to pack in more distortion products and so uses up more RF power.
So the question I have is several fold:
1. apparently these distortion products are harmless, at least in the noisy environment of HF radio, is this in fact true?
2. does packing in these additional distortion products really help you punch through to the other side?
3. would non-clipping baseband processing be just as much, less, or more, effective than forcing a controlled clipping with either of "fast acting ALC" or CESSB?
It might just be the case that the CESSB algorithm is a "better" (?) way of controlled clipping, without the attack and recovery periods found in an ALC system.
A lot of the answer really depends on how we hear speech, and can't be answered by making linear measurements. So I'm really asking people who have used CESSB how their experience with it has gone?
73 de Dave, N7AIG
Yes, I did a bit more digging last night and found the Tong paper. In there he shows a graph that describes how the peaks in voiced audio occur only rarely, compared to the more prominent vowel phonemes. So any distortion is likely to be very short lived, and these processing schemes do achieve some substantial boosting of the vowels.
But I also found another paper by Leif, SM5BSZ
In there he shows some very interesting results. For one thing, it appears that all current transceiver manufacturers fail to heed the knowledge provided back in the 1980's by Sabin, et al, wherein he describes the need for a dual ALC system: one very fast acting with 0.1 s recovery, and only 3 dB range, to handle the re-peaking that occurs in SSB when you re-filter the processed RF to remove IMD and harmonics caused by clipping, and another longer acting one, wide dynamic range, to manage overall power levels with respect to temperature, aging, power settings, etc.
In particular he shows that re-peaking must occur as a result of post processing filtering, since you can only be flat-topped if you hold onto all the harmonics produced by clipping.
But back to CESSB... that avoids overt clipping, per se, but surely produces distortion products within the audio pass band. Again, these would be short lived peaks. You can see that this is so by simply examining his envelope shapes after processing. There is a predistortion of the envelope to avoid the production of clipping products.
While Leif shows the benefit overall of RF clipping, or intentional distortion, CESSB appears to be a way to overcome the lack of the fast acting ALC. Interestingly, RF clipping produces essentially zero distortion on pure tones, pairs of tones, and glottal (Guassian) pulses. CESSB will go ahead and distort these waveforms too.
Leif also provides the kind of data that you really need to see... namely, a measurement of speech recognition effectiveness versus SNR, for a variety of processing methods. And without doubt, (not comparing CESSB) he shows that 15-20 dB of RF clipping compression is the most effective overall, possibly coupled with audio compression to even out 10 dB of mic-speaker distance and gross speech level variations.
But it is interesting to hear the comments of phone ops. I almost never talk over the air, so thanks for your feedback!
73 de Dave, N7AIG
BTW... along the way, I ran across a little gem for RX of SSB voice. It seems to work pretty well to cut down on trash noise in the passband.
Voices typically exhibit a trough between 700 Hz and 1400 Hz, which lies between the main formants of speech. So notching out that region removes incoming noise that carries little intelligence.
- de N7AIG
Using ESSB for DX is just poor radio practice. Why Because the majority of DX stations receive no larger than 2.8 wide Most listen Far smaller than that . (Shorten that Band pass to 2.8 and add the DX+ see the real difference)
Here is the real deal. Fixed radio compression or One adaptive process containing the Audio spectrum is not the solution. Granted this is what the radio compression levels give us.
If you are going to project your SSB audio (LOUD) by adding compression The correct way to attain the desired or best results is to stage the events by various frequencies within your audio band pass .
YEP not one lump constant but the sum of compression indexes along the audio frequencies that will Add to the total audio package .
Now most of that increased IMD is derived from LOW frequency enhancement .
If you don't believe just reduce that 700 HZ tone in your 2 tone test and see the difference in the IMD .
Since I am now working DX stations running Loud audio , Also clear signal audio I can tell you all that You want more DX put that power into the band pass .
By the way I have not found any other way to accomplish this but with Electronic audio processing .
Flex you want to lead the pack put adaptive frequency control on compression . Add slope , Q ., and +/- gain . WOW now were talking